Prediction models that capture and use the structure of state-space dynamics can be very effective. In practice, however, one rarely has access to full information about that structure, and accurate reconstruction of the dynamics from scalar time-series data-e.g., via delaycoordinate embedding-can be a real challenge. In this paper, we show that forecast models that employ incomplete embeddings of the dynamics can produce surprisingly accurate predictions of the state of a dynamical system. In particular, we demonstrate the effectiveness of a simple near-neighbor forecast technique that works with a two-dimensional embedding. Even though correctness of the topology is not guaranteed for incomplete reconstructions like this, the dynamical structure that they capture allows for accurate predictions-in many cases, even more accurate than predictions generated using a full embedding. This could be very useful in the context of real-time forecasting, where the human effort required to produce a correct delay-coordinate embedding is prohibitive.
LEAD PARAGRAPHPrediction models constructed from state-space dynamics have a long and rich history, dating back to roulette and beyond. A major stumbling block in the application of these models in real-world situations is the need to reconstruct the dynamics from scalar time-series data-e.g., via delay-coordinate embedding. This procedure, which is the topic of a large and active body of literature, involves estimation of two free parameters: the dimension m of the reconstruction space and the delay, τ , between the observations that make up the coordinates in that space. Estimating good values for these parameters is not trivial; it requires the proper mathematics, attention to the data requirements, computational effort, and expert interpretation of the results of the calculations. This is a major challenge if one is interested in real-time forecasting, especially when the systems involved operate a) Electronic mail: joshua.garland@colorado.edu b) Electronic mail: lizb@colorado.edu on fast time scales. In this paper, we show that the full effort of delay-coordinate embedding is not always necessary when one is building forecast models, and can indeed be overkill. Using synthetic time-series data generated from the Lorenz-96 atmospheric model and real data from a computer performance experiment, we demonstrate that a two-dimensional embedding of scalar timeseries data from a dynamical system gives simple forecast methods enough traction to generate accurate predictions of the future course of those dynamics-sometimes even more accurate than predictions created using the full embedding. Since incomplete embeddings do not preserve the topology of the full dynamics, this is interesting from a mathematical standpoint. It is also potentially useful in practice. This reduced-order forecasting strategy involves only one free parameter (τ ), good values for which, we believe, can be estimated 'on the fly' using information-theoretic and/or machine-learning algorithms. As suc...